Bloomberg Intelligence Puts AI Data Center Build Cost at $40 Billion to $60 Billion Per Gigawatt
The capital cost estimate highlights grid access as the binding constraint on AI infrastructure expansion, with BNEF tracking 100 GW of U.S. project pipeline added in the past year.
Building one gigawatt of AI data center capacity costs between $40 billion and $60 billion, Bloomberg Intelligence estimated. Power procurement and grid interconnection account for the bulk of that cost. The same Bloomberg Intelligence analysis reported that Microsoft is planning to triple its computing power, with CapEx ramping sharply toward what analysts described as gigawatt-scale expenditure levels.8
The AI data center market is now in a phase where capital is available and customers are willing to sign long-term capacity contracts, DataM Intelligence noted, but power availability and interconnection delays remain the chief deployment bottleneck. That gap between capital readiness and grid access is pushing developers toward non-utility power solutions years ahead of planned operations.4
BloombergNEF tracked approximately 100 GW of new U.S. data center project capacity added in the past year, Senior Associate Nathalie Limandibhratha said on Wednesday (2026-07-22). Almost all U.S. regions ended 2025 with more data center capacity than BNEF had anticipated. Texas showed the sharpest gap between forecast and actual buildout.5
Demand forecasts carry wide uncertainty. BNEF's two main scenarios for U.S. data center electricity demand by 2030 differ by 42 GW, reflecting competing assumptions about chip efficiency, model architecture, and deployment pace. The firm's chip-based model puts U.S. total demand at 207 GW by 2033. BofA analysts have separately estimated the potential U.S. incremental load addition at around 125 GW. BNEF described its base case as sitting in the middle of third-party estimates.5
McKinsey estimates AI-related infrastructure spending could exceed $5 trillion by 2030, with JLL projecting roughly 100 GW of new data center capacity required over the same period. DataM Intelligence put the global AI data center market at $120.74 billion in 2025, forecasting growth to $1,020.83 billion by 2035 at a compound annual growth rate of 22.8%, driven by generative AI adoption.3,4
Virginia shows what this demand profile does to individual power markets. Commercial electricity sales there rose by nearly 30 million megawatt-hours between 2019 and 2025, with the EIA attributing much of that increase to the state's dense concentration of data centers. The U.S. now accounts for nearly 40% of global data center electricity consumption, Forbes reported in August (2026-08-23).6,7
Unable to wait for utility interconnection, developers are pursuing unconventional supply. Crusoe signed a $1.25 billion contract with Boom Supersonic for 29 jet-engine turbines to power data centers around the country. Startup Panthalassa raised $140 million for offshore buoy-mounted facilities that would generate power from wave energy. The Atlantic Council flagged that some AI inference operations are already moving toward surge pricing during peak demand periods, with operators under pressure to treat data centers as flexible load assets rather than static baseload consumers.2,1
Gas demand from the buildout is real but limited in aggregate. The Atlantic Council estimated that even a 4-6 gigawatt facility consumes only around one billion cubic feet per day of natural gas, depending on turbine efficiency — material locally but modest against national supply balances. NYMEX Henry Hub front-month held at $2.91/MMBtu on Wednesday (2026-09-16), showing no market response to the data center demand narrative.1
The 42-GW spread between BNEF's two demand scenarios keeps power generators, gas suppliers, and transmission developers uncertain on capacity decisions. Texas has consistently outpaced prior BNEF forecasts, suggesting the lower demand case is already conservative. But the upper bound hinges on chip efficiency improvements stalling out, and that has not happened yet. Grid operators in Texas and PJM revising their interconnection timelines in response to the project pipeline BNEF is now tracking will be the clearest early indicator of how utilities plan to absorb the load.5